The arc plasma torch can be used for pre experiments on ground erosion performance testing of spacecraft flight materials，which can save costs. The three-phase AC arc plasma torch has the advantages of simple power supply and reliable operation. The hollow electrode structure with dual inlet channels can not only improve the electrode life, but also achieve a wider range of power control. However, the design of plasma torches with this type of electrode structure is more complex and there is limited research and application in China. A three-phase AC plasma torch with magnetic motion, tangential inlet, and supersonic jet was developed and numerically modeled and experimentally studied.
Firstly, a three-dimensional turbulent MHD multiphysics coupling simulation model of a hollow electrode three-phase AC arc plasma torch with a dual end inlet structure was established, and the flow state and electric thermal characteristics of the arc plasma inside the torch were obtained. Secondly, the influence laws of air intake, working current, air intake distribution ratio, and working frequency on the electric field, magnetic field, temperature field, flow field distribution, and arc characteristics inside the plasma torch were studied and revealed. Finally, the correctness of the numerical model was verified by comparing the arc voltage, nozzle outlet temperature, and arc root position under various operating conditions in simulation and experiment.
The conclusion drawn from the study is as follows: (1) In a three-phase AC plasma torch, aerodynamic and electromagnetic forces dominate the flow characteristics of the arc root. During the process of increasing the intake volume from 30 g/s to 60 g/s, the cooling effect of the gas flowing along the wall is greater than the heat generated by the arc column, resulting in a downward trend in temperature; And the larger the intake volume, the more obvious the compression effect of the cold air layer on the arc, and the higher the arc pressure; The higher the working current, the higher the plasma temperature and jet velocity. (2) In a hollow electrode AC plasma torch with dual inlet ducts, changing the air intake distribution ratio can alter the position of the arc root along the electrode axis and the magnitude of the output power. Increasing the air intake distribution ratio can make the arc more significantly stretched in the axial direction, the arc longer, and the arc root closer to the arc back cover. (3) When the operating frequency is 1 kHz, the arc has a more stable motion trend, and the rotation speed of the arc root is five times that of the power frequency. The contact area with the electrode is reduced, which reduces the degree of electrode erosion and can improve the electrode life.

As electric vehicles (EV) grow more popular and vehicle-to-grid (V2G) technology advances, large-scale EV aggregations (EVA) have become integral to the power system. However, effectively capturing the distinct idle energy storage characteristics of EVAs across regions and integrating them seamlessly into power system operations remains a challenge. The shortcomings of existing research can be summarized as the follows: Firstly, current methods for assessing the dispatchable regions (DR) of EVs remain inadequate, lacking systematic frameworks and classification methods. Secondly, current multi-level coordinated control strategy often overlooks the holistic nature of coordinated control, which spans multiple levels, including the power grid, garage, and users. Merely considering factors related to EVs and their users is insufficient, as it fails to provide a comprehensive guidance for all coordinated control participants, such as the power grid and garage.
This paper addresses the aforementioned issues by conducting the following works. Firstly, methods for establishing multi-stage electric vehicle dispatchable region (MEVDR) for both EV and EVA are proposed and further investigated. Secondly, the probability density functions of various EV data in different regions and time periods of clustering centers are captured using Gaussian mixture model (GMM). Thirdly, the MEVDR of EVAs in different regions and time periods are established and comprehensively analyzed. Furthermore, the proposed MEVDR model can be used to construct multi-period constraints. Based on this, a vehicles-garage-grid multi-level coordinated control system (VGGMCCS) based on MEVDR can be constructed, which consists of two levels and can therefore be considered a bi-level model. After a thorough analysis, VGGMCCS incorporates two mixed integer programming (MIP) problems, allowing the use of commercial solvers for rapid and efficient problem solving. Finally, in order to provide further validation of the effectiveness of the VGGMCC system based on MEVDR, a comparison was made between the proposed method and the contrasting strategies.
The case study shows that, when compared to two contrasting strategies, the proposed VGGMCCS has been demonstrated to reduce the grid network loss by 12.17% compared to comparative strategy 1 and by 8.69% compared to comparative strategy 2 during peak electricity demand periods. And to reduce users′ average daily charging costs by 7.88% compared to comparative strategy 1, and to increase operators′ revenues by 17.63% compared to comparative strategy 2. Meanwhile, the load fluctuation amplitude of the transformer at the garage node has been significantly reduced. During peak electricity consumption periods, the power fluctuation of transformers under VGGMCCS decreased by 96.36% compared to comparative strategy 1 and by 82.59% compared to comparative strategy 2. Last but not least, VGGMCCS also has a high solution speed, ensuring decision accuracy while quickly responding to dispatching requests from lower-level garages, effectively reducing both the time and economic losses caused by rescheduling requests after EVs are integrated into the power grid. The results show that VGGMCCS can effectively reduce users′ costs, improve the economic benefits of the garage and enhance the operational efficiency of the power grid, while ensuring the long-term stable operation of the power system, thus achieving a win-win situation for users, garage operators and power grid companies.
In summary, this paper provides a thorough establishment and analysis of EVA′s MEVDR across a diverse range of geographical and temporal contexts. Furthermore, when compared to the contrasting strategies, the proposed VGGMCCS promises to enhance both the economic benefits and operational efficiency of the power system significantly.

Metallic foreign objects in various types of power equipment may cause discharge problems. To achieve accurate multi-spectral monitoring of the discharge phenomenon caused by metallic foreign object, it is necessary to deeply understand its influence on the mechanism of optical radiation during discharge. Currently, research on the impact of metallic foreign objects on high-voltage discharge is mostly focused on the macro level, without delving into the micro-particle level to analyze its effect on the discharge mechanism, and the influence of metallic foreign object on optical radiation during discharge has not been thoroughly explored. To address these issues, this paper analyzes the characteristics of metallic foreign object's impact on the full-band optical radiation of discharge and its influence mechanism on the day-blind ultraviolet band through experiments and simulations.
A high-voltage discharge experimental platform was first constructed. Discharge images were captured using ultraviolet and high-speed cameras, and the emission spectra were measured with a spectrometer to investigate the influence of metallic foreign object on full-band optical radiation during discharge. The effect of metallic foreign objects on the generation of day-blind ultraviolet radiation was further studied. It was verified that the particle transitions responsible for producing day-blind ultraviolet radiation are mainly $\mathrm{N}_{2}\left(\mathrm{~A}^{3} \Sigma_{\mathrm{u}}^{+} \rightarrow \mathrm{X}^{1} \Sigma_{\mathrm{g}}^{+}\right)$ and NO-γ(A²Σ⁺(v′)→X²Π(v″)). Based on this, a two-dimensional plasma simulation model was constructed to investigate the effect of different quantities of large metallic particles and varying masses of metal shavings on the discharge process. The model was used to calculate the number densities of NO(A²Σ⁺) and $\mathrm{N}_{2}\left(\mathrm{~A}^{3} \Sigma_{\mathrm{u}}^{+}\right)$ particles under different conditions, and the simulation results were validated by comparing them with the measured spectra. The experimental and simulation results were then comprehensively analyzed to explore the influence of metallic foreign object on day-blind ultraviolet radiation during discharge.
High-speed camera reveals that metallic foreign object increases the chance of arc formation between the tip of the needle electrode and the metallic foreign object. From the spectrum of the 200-1000 nm band measured in the experiment, it is evident that the increase in metallic foreign object enhances the optical radiation across the entire spectrum generated by the discharge. However, this enhancement is selective to certain bands, with the ultraviolet and visible light bands responding more sensitively. Therefore, ultraviolet and visible light detection is more effective for monitoring discharges caused by metallic foreign objects.
Analysis of UV images, 240~280 nm spectra, and simulations shows that an increase in the metal foreign object causes an increase in the amplitude of the spectral curve of the sun-blind UV band, an increase in the number densities of NO(A²Σ⁺) and $\mathrm{N}_{2}\left(\mathrm{~A}^{3} \Sigma_{\mathrm{u}}^{+}\right)$ particles, and an increase in the rate of the chemical reactions in the discharge region; however, the spectral shape remains basically unchanged, which means that it does not affect the types of chemical reactions and the relative ratios among them. By combining the electric field simulation results, the reason can be analyzed as follows: the metal foreign object increases the strength and inhomogeneity of the electric field, promoting the excitation and ionization of particles. This leads to the production of more NO(A²Σ⁺) and $\mathrm{N}_{2}\left(\mathrm{~A}^{3} \Sigma_{\mathrm{u}}^{+}\right)$ particles, thus promoting the enhancement of the sun-blind ultraviolet radiation.
The results of this paper apply to discharge phenomena in air influenced by metallic foreign objects, and the influence of metal particles on discharge in SF₆ and its alternative gases will be further investigated in the future.

With the development of wind power research, the factors considered in the simulation model are gradually increasing. The demand for wind turbine models that take into account both electrical and mechanical characteristics is increasing, thus promoting the development and application of co-simulation technology. However, the detailed electrical model has strict requirements on the simulation time step, which reduces the efficiency of co-simulation. Some scholars have properly simplified the electrical model to improve the simulation speed when carrying out co-simulation. But, improper selection of simulation time step has a negative impact on simulation accuracy. Therefore, this paper studies the model optimization and step size selection to solve the problem of the contradiction between simulation accuracy and speed.
First, the complex electrical model is optimized by ignoring the power electronic switching model and reducing the order of the higher-order model, so that the computational complexity is reduced and the application range of the simulation step size is increased. GH Bladed and Matlab/Simulink are selected to build the co-simulation platform. Then, a comprehensive evaluation method based on residual similarity and feature selection verification is proposed, which takes into account simulation accuracy and speed. The evaluation of simulation accuracy is divided into two aspects: global and transient difference. The comprehensive evaluation index is formed by combining the evaluation index of simulation accuracy and simulation speed in a weighted way to guide the selection of simulation time step. Finally, the co-simulation is carried out to verify the effect of the optimization model on the simulation efficiency under the conditions of wind speed disturbance, frequency disturbance and fault crossing disturbance. According to the simulation results and the comprehensive evaluation method, the reference suggestions for the selection of simulation step size are put forward.
Through simulation results and analysis, the following conclusions are drawn: (1) Through the optimized model, the co-simulation model can be run at a larger simulation time step, and the co-simulation efficiency is improved. (2) According to the proposed comprehensive evaluation method, the simulation results are evaluated from two dimensions of simulation accuracy and speed, which solves the problem of quantitative evaluation of the accuracy and speed of the model simulation results. (3) Through the co-simulation of various working conditions and the quantitative evaluation of simulation accuracy and speed, the following conclusions are drawn: Under the background of the simulation in this paper, under the condition of wind speed fluctuation, the simulation time step of co-simulation is chosen to be around 0.05 s; under the condition of frequency disturbance, the simulation time step of co-simulation is chosen to be around 0.01 s; under the condition of fault ride-through disturbance, the simulation time step of co-simulation is chosen to be around 0.005 s. (4) Based on the evaluation results of multi-condition simulation, the factors such as time scale of simulation condition and mutation characteristics of observed parameters should be fully considered in the selection of simulation time step. When the time scale is large and the observed parameters do not have mutation characteristics, the larger simulation time step can be selected. When the time scale is small and the observed parameters have mutation characteristics, the selection of simulation time step should be reduced appropriately.

In recent years, with the in-depth reform of the electricity market and the continuous improvement of the penetration rate of distributed resources in the distribution network, the energy trading between intelligent buildings with dual attributes of production and consumption has brought new opportunities and challenges to the nearby consumption of distributed energy. However, for the microgrid system with multi-intelligent buildings, there are defects such as large amount of communication information, low robustness and user privacy in the process of power trading. At the same time, it will also be affected by various uncertain factors such as the access of new energy and the lack of timeliness of transactions. In order to solve the above problems, this paper proposes a rolling P2P energy trading optimization strategy based on distributed information interaction for multi-intelligent buildings in microgrid.
Firstly, considering the aggregation characteristics of various flexible resources in intelligent buildings, the prediction interval results of distributed photovoltaic power generation and the feasible range of flexible resources are characterized in the form of aggregation power interval by Minkowski summation theory, and the aggregation interval model of P2P transaction is established. Among them, the distributed photovoltaic prediction interval is modeled by transforming the benchmark output at different confidence levels into the prediction quantile for the feasible region. At the same time, an interval rolling P2P energy trading framework is constructed. During the energy management period, each building participates in the rolling P2P energy trading by combining the aggregated power interval with its own electricity purchase and sale strategy. Secondly, the risk cost brought by the uncertainty of photovoltaic output to P2P transactions is quantified by CVaR, and an economic dispatch model with the minimum total operating cost of microgrid multi-intelligent buildings is established. On this basis, the P2P transaction power between buildings is used as a consistency variable, and the P2P transaction power and transaction price are obtained based on the distributed solution of the information interaction between adjacent buildings, and the energy transaction period is continuously pushed backward until it meets the requirements of all intelligent buildings in the microgrid.
In the case analysis, the scheduling results of different buildings in the microgrid and the optimization results of different algorithms are compared respectively, which verifies the effectiveness of the interval rolling P2P energy trading model proposed in this paper. At the same time, the practicability and solution efficiency of the distributed information interaction algorithm in this paper have also been reflected. Through the example analysis, the following conclusions can be drawn: (1) Participating in the energy transaction between buildings in the form of aggregation interval fully taps the scheduling potential of flexible resources in buildings and improves the flexibility of coordinated scheduling of multi-intelligent buildings in microgrid. (2) Compared with the ordinary P2P trading, the rolling P2P energy trading improves the enthusiasm of intelligent buildings to participate in energy trading and the self-consumption level of distributed energy while taking into account the economy of system operation. (3) The distributed information interaction strategy proposed in this paper makes the multi-intelligent buildings in the microgrid only need to interact with the expected transaction volume information, and at the same time solve their own optimization problems in parallel, which has a good fit with the rolling P2P transaction mode. It avoids the problems of high computational pressure and privacy leakage, and improves the convergence speed of distributed information interaction. It has good scalability and can effectively solve the optimization iteration problem of large-scale intelligent buildings.

Epoxy resin is widely used in epoxy cast electrical equipment such as dry-type transformers and dry-type reactors due to its good mechanical strength, chemical corrosion resistance, and excellent electrical insulation performance. However, the irreversible cross-linking network formed after curing makes it difficult to degrade and recycle retired electrical equipment. Researchers have developed a series of biodegradable resins with high electrical thermal mechanical properties and degradation characteristics by introducing dynamic covalent bonds. However, epoxy electrical equipment such as dry-type transformers and dry-type reactors that operate in complex environments such as high temperature, high electric field, and mechanical vibration for a long time can experience performance degradation due to resin aging, which affects their service life. The changes in the cross-linking structure of epoxy resin caused by thermal oxidative aging may have a certain impact on the service performance and degradation recovery characteristics of degradable resins. This article used the ester exchange catalyst triethanolamine to construct a degradable epoxy resin system, and conducted accelerated thermal oxidative aging tests on it to analyze the effects of aging time and catalyst on the service performance and degradation characteristics of degradable epoxy resin.
Firstly, this article used ester exchange catalyst triethanolamine to construct a degradable epoxy resin system, and used traditional non degradable epoxy resin as a reference to conduct thermal oxidative aging tests on resins with different triethanolamine contents at three temperatures of 180℃, 200℃, and 220℃. Then, the performance changes of different resin systems after aging were studied through comprehensive analysis of electrical properties, thermogravimetric analysis, dynamic thermomechanical analysis, mechanical properties and microstructure analysis. The bending strength retention rate was used as an aging index to estimate the service life. Finally, this article also explored the influence of thermal oxidative aging on the degradation properties of degradable resins.
From the experimental analysis, the following conclusions can be drawn: (1) The insulation and electrical performance of the degradable epoxy resin system after high-temperature aging is slightly worse than that of traditional resins, but the degradation rate of the insulation performance of degradable resins is slower than that of traditional resins under 200℃ and 220℃ conditions, with V-TEOA-0.05 maintaining better electrical performance. (2) The thermal stability of the degradable epoxy resin system is slightly inferior to traditional resins, but V-TEOA-0.05 has a higher storage modulus and a slightly lower glass transition temperature, and also exhibits good thermal properties. (3) As the aging temperature increases, the difference in flexural strength between degradable epoxy resin and traditional resin after aging gradually narrows, and remains basically unchanged after 49 days of aging at 220℃. The estimated lifespan of the V-TEOA-0.05 system shows a temperature index of 163.13℃, demonstrating excellent heat and oxygen aging resistance. (4) In the mixed solution of EG and TBD, the degradation rate of V-TEOA-0.05 sample decreases with increasing aging time, which may be related to the increase in resin crosslinking density, decrease in free volume, and decrease in ester bonds.

Applying the vacuum switch in the more-electric aircraft intermediate frequency (IF 360~800 Hz) power system is a new application field, which can solve the difficulties caused by the increase of current frequency and the limited breaking ability of electrical appliances. The anode activity of the vacuum arc determines the post-arc state and interruption ability of the vacuum switchgear, especially at high current, and the anode can actively emit metal vapor, plasma and metal droplets. Because of the special environment of the vacuum chamber, it is difficult to directly measure the physical quantity of the post-arc state, such as arc pressure, by using the sensor, so non-contact measurement means is generally adopted. To gain a more comprehensive understanding of the post-arc characteristics of intermediate frequency vacuum arcs, the visual tracking techniques such as object detection and Intersection over Union Tracker were utilized to analyze arc images in this paper. The splatter trajectories of post-arc metal droplets were reconstructed in three dimensions. Based on the reconstruction, the spatial pressure gradient inside the arc was determined.
Firstly, an intermediate frequency vacuum arc experimental system was established, along with a dual high-speed camera stereoscopic arc imaging system. Secondly, the experimental results of the intermediate frequency vacuum arc were analyzed, revealing post-arc voltage oscillations and metal droplet ejection phenomena during interruption failure. Thirdly, utilizing visual tracking techniques such as Canny edge detection, connected component analysis, and IoU, along with the mapping relationship from arc plane to three-dimensional space, a method for analyzing the pressure gradient of the post-arc vacuum arc was developed. The detection and tracking performance of arc images were evaluated using metrics such as precision, recall, MOTA, and MOTP, achieving values of 91.69%, 84.28%, 87.19%, and 82.63%, respectively, indicating excellent visual tracking results. Finally, using the aforementioned theories and methods, a comprehensive analysis of the post-arc characteristics of the intermediate frequency vacuum arc was conducted.
The following conclusions can be drawn from the analysis: (1) According to experimental results, when post-arc breakdown occurs after the intermediate-frequency current crosses zero, the arc voltage exhibits high-frequency oscillations with a frequency of approximately 50 kHz. The voltage stabilizes within about 2 ms. During the post-arc period, dual-view arc images reveal substantial outward ejection of metal droplets. (2) By employing visual tracking algorithms and spatial mapping relations, the three-dimensional ejection process of metal droplets during the post-arc breakdown can be reconstructed. The acceleration in all three directions reaches the order of 10⁵ m/s², with ejection velocities on the order of 10 m/s. The pressure gradient within the arc chamber can reach 1.2 MPa/mm, and the time scale for droplets to travel from the contact edge to the inner wall of the arc chamber is milliseconds. (3) The vapor density of Cu on the surface of the metal droplets is 2.2×10¹⁹ m^-3. Throughout the ejection process of milliseconds scale, the metal droplets continuously evaporate, reducing the Cu mass fraction on the droplet surface from 65% to 10%. A significant amount of Cu vapor enters the arc chamber through diffusion and convection, weakening the dielectric recovery strength post-arc. During this period, post-arc breakdown and high-frequency voltage oscillations occur.

Addressing the issues of inadequate exploitation of hydrogen energy collaboration potential and the challenge in balancing accuracy and efficiency of probabilistic solution algorithms, this paper proposes a calculation method for the probabilistic optimal energy flow of electricity-hydrogen systems based on compressed sparse arbitrarily polynomial chaos expansions (CS-aPCE).
Firstly, to harness the spatial-temporal collaboration potential of hydrogen energy, a modeling approach for electricity-hydrogen optimal energy flow is introduced incorporating peer-to-peer (P2P) hydrogen collaboration between on-site and off-site hydrogen refueling stations (HRSs). Considering the P2P coupling of inter-station hydrogen flows and bidirectional electricity flows, a P2P collaboration mechanism is proposed for between on-site and off-site HRSs and the power distribution network. Based on Nash bargaining theory, a probabilistic optimal electricity-hydrogen energy flow model is constructed, which incorporates time-delay and discreteness constraints for inter-station hydrogen P2P transactions. This model coordinates multi-stakeholder benefit allocation and electricity-hydrogen price decisions, enhancing feasibility and fairness.
Secondly, a probabilistic optimal energy flow solution algorithm for on-site and off-site HRSs and power distribution networks is proposed based on CS-aPCE, aiming to improve the efficiency and accuracy of high-dimensional probability calculations. The core of this algorithm lies in leveraging historical data to drive the collocation points, subsequently calculating key statistical metrics such as expectations and standard deviations through analytical methods, without reliance on prior probabilistic information. To further optimize computational performance, the CS-aPCE algorithm integrates Gaussian quadrature rules to construct high-frequency collocation points and incorporates compressed sparse grid techniques. Effective compression criteria are proposed, and the dimensionality reduction effect and computational accuracy of the algorithm are theoretically proven, ensuring its efficiency and robustness under high-dimensional randomness.
The effectiveness of the proposed method is validated through numerical examples, leading to the following conclusions: firstly, the CS-aPCE algorithm presented in this paper can solve high-dimensional and probabilistic electricity-hydrogen energy flow problems rapidly and with high precision. The computation time is merely 10% of that required by the Monte Carlo simulation method, while the errors in expected values and standard deviations are below 4.21%. Furthermore, the computational accuracy for higher-order moments is improved by 60.28% to 156.98% compared to the traditional aPCE method. Secondly, the stationarity threshold exerts a certain influence on the accuracy and efficiency of the CS-aPCE algorithm. A reasonable threshold should be selected by comprehensively considering the stationarity distribution characteristics of random variables. Finally, the electricity-hydrogen optimal energy flow model considering P2P hydrogen collaboration between stations can mobilize the coordination potential of flexible resources within the distributed hydrogen supply network, coordinate the distribution of inter-station hydrogen flows, and achieve fair allocation of benefits among multiple stakeholders.

Internal short circuit is one of the most serious faults in transformers, which can lead to a rapid increase in fault energy in a short period of time and easily cause high-energy discharge and explosion inside the equipment. However, there are many potential combinations of internal short circuit conditions in transformers. The analysis method of field-circuit coupling commonly used by transformer manufacturing enterprises has the problems of excessive time and resource consumption. And it is difficult to model jointly with the external power grid. Existing circuit models face difficulties in multi-scale coupling characterization and parameter calculation of windings.
This article focused on the urgent need for transformer short circuit fault analysis. A construction method of multi-scale fault analysis model for single-phase transformer with internal short circuit was proposed. Firstly, based on the multi-scale characteristics of transformer windings and internal short circuit faults, the transformer windings were virtually divided into several sub-windings using axial segmentation. By parametrically scanning the finite element model of the transformer, the self-mutual inductance matrix and resistance matrix of sub-windings was calculated. Secondly, an calculation method was proposed to transform the self-mutual inductance matrix into the coupled leakage inductance matrix, which could effectively characterize the leakage magnetic characteristics between sub-windings. This parameter calculation method could be carried without port short circuit tests, which solved the problem of parameter calculation for existing multi-winding transformer models. Finally, a multi-scale circuit model for transformers was established based on the coupled leakage inductance matrix. By connecting the terminals of each sub-winding based on the electromagnetic connection relationship and the physical process of internal short circuit, transformer fault analysis models for different internal short circuit conditions could be obtained. The problems of low efficiency and poor circuit adaptability in the fault analysis model based on field-circuit coupling were solved.
Furthermore, a disk-scale circuit model of an 80 MV·A single-phase transformer was constructed. A comparative simulation was conducted with the finite element model. The results indicated that the errors of the short-circuit impedance and the peak value of the port current at rated operating condition were almost zero. And the simulation time was reduced by about 99.98%. After single inter-turn short circuit faults, the errors of the first peak values of the port currents and short-circuit currents did not exceed 2.5%. The simulation efficiency was improved while ensuring simulation accuracy. Then, based on a certain engineering accident, a developmental inter-turn short circuit analogy simulation analysis was carried out. The errors of the first peak values of the port currents and short-circuit currents after the fault, as well as the local peak values during the fault development process, did not exceed 5.5%. And the duration of the second harmonic percentage of fault differential current accounting for more than 15% of the circuit model was calculated to be 49 ms. It was consistent with the finite element model calculation results. The existing method was 14ms. Therefore, the proposed construction method of multi-scale fault analysis model for single-phase transformer with internal short circuit can accurately simulate the transient characteristics of transformers with internal short circuit under multiple scales and operating conditions. This method provides a basic model for research on equipment accident analysis, traceability, and fault defense.

Power battery packs are widely used in new energy electric vehicles and are the core components of electric vehicles. Studying the temperature field modeling of the power battery pack is not only beneficial to understanding its temperature field dynamic characteristics, but is also very important for the structural design and health management of the power battery pack. The temperature field of the power battery pack is described by complex partial differential equations. Since a large number of parameters are unknown and many model parameters show strong time variability, traditional physics-based modeling methods are ineffective in achieving online modeling of the temperature field of the power battery pack. Although methods based on deep learning do not rely on physical models, they require a large amount of experimental data during the training process, the model training time is long, and the real-time performance of temperature field prediction is poor. In response to the above problems, this paper proposes a spatio-temporal modeling of the temperature field of power battery packs based on long short-term memory network.
First, the spatio-temporal separation method is used to extract spatial features and time features under offline conditions. Spatial features are continuously updated with the help of incremental learning, and the long short-term memory (LSTM) network is used to model temporal dynamics. Finally, the updated spatial characteristics and time model are integrated to obtain a prediction model of the power battery pack temperature field.
The proposed method was verified on a power battery pack composed of 24 battery cells. Experimental results show that the proposed method can accurately predict the temperature field of the power battery pack regardless of normal conditions or conditions with air flow interference. Without airflow interference, the single-point temperature prediction error of the proposed method is less than 0.4℃, and the root-mean-square error (RMSE) on the test set is 0.095 1℃. In the presence of airflow interference, the single-point temperature prediction error of the proposed method is less than 0.07℃, and the RMSE on the test set is 0.014 7℃.Under the condition of air flow, the modeling error of the proposed method is smaller. This is because under the condition of air flow interference, the spatial gradient of the temperature change of the power battery pack at the same time is smaller, that is, the temperature change is gentler, making the spatial characteristics of the modeling smoother.
The following conclusions can be drawn from the simulation analysis: (1) the proposed method can accurately predict the temperature field of the power battery pack regardless of normal conditions or conditions with air flow interference. (2) The proposed method can update spatial features in real time through incremental learning, thereby reducing the computational complexity of the method. (3) The proposed method is a purely data-driven method that does not rely on accurate partial differential equations and is therefore suitable for application in temperature field modeling of actual power battery packs.